Imagens multiespectrais no monitoramento de parâmetros morfobiométricos do cafeeiro sob tratamentos químicos e biológicos contra fitonematoides

• Ano de defesa: 2021
• Autor(a) principal: Pereira, Fernando Vasconcelos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Agricultura e Informações Geoespaciais
• Palavras-chave em Português: Sensoriamento remoto; Remote sensing; Parâmetros agronômicos; Agronomic parameters; Aprendizado de máquina; Machine learning; Nematoides; Nematodes; Agronomia; Agronomy; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::GEOFISICA::SENSORIAMENTO REMOTO; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::FITOSSANIDADE::FITOPATOLOGIA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/32727; http://doi.org/10.14393/ufu.di.2021.458

Coffee plays an important role in the Brazilian economy. Much like other crops, coffee is exposed to different pathogens and pests that directly affect yield. These include nematodes, which attack the roots of plants and compromise their physiological development. Given the losses caused by this pathogen and the lack of information on spatial distribution in infested areas, it is important to adopt technologies that enable crops under different management systems to be monitored during their growth cycle in order to reduce nematode populations. In this respect, remote sensing associated with machine learning algorithms is presented as a potential tool for monitoring agricultural crops using multispectral images. The present study assesses different machine learning algorithms, using radiometric values of multispectral images obtained by remote sensing platforms as input datasets, and identifies the best architectures (Random Forest, Multilayer Perceptron, SMOreg and Linear Regression), input datasets (spectral bands, vegetation indices, and combination of the two, selected in cluster analysis) and remote sensors (RPA, MAPIR and the PLANET platform) to estimate the agronomic parameters of yield in coffee crops submitted to 11 treatments for nematode management. The best-performing architectures were those that obtained the lowest RMSE and RMSE% values, as follows: total chlorophyll index (Random Forest/ vegetation indices/ RPA) with respective RMSE and RMSE% of 4.7975 and 9.0545; plant height (m), branch length (south-facing) (m) and branch length (north-facing) (m) (SMOreg/ bands and vegetation indices selected by cluster analysis / RPA), with respective RMSE and RMSE% of 0.1128 and 3.6929; 0.1329 and 15.3025; 0.1436 and 16.8162; number of branches and number of nodes (south-facing), (Linear Regression/ spectral bands and vegetation indices selected by cluster analysis/ RPA), with RMSE and RMSE% of 12.1711 and 16.4744; 5.0442 and 18.2725; number of nodes (north-facing) (Random Forest/ spectral bands/ PLANET), with RMSE and RMSE% of 7.5341 and 26.2917; canopy diameter (m) (SMOreg/ spectral bands/PLANET), with RMSE and RMSE% of 0.1302 and 7.7374.